What the data for this use case looks like
Data sources for semiconductor wafer fabrication production ledgers include MES systems, equipment FDC modules, defect detection and metrology equipment, wafer map files, process recipe libraries, and experimental datasets. Data update frequencies fall into three categories: triggered per wafer batch transfer, real-time equipment parameter reporting, and daily aggregate statistics. Daily documents include process flow cards, batch operation records, equipment recipes, SPC analysis charts, exception handling sheets, defect maps, and yield reports. Fields include nanoscale process parameters, batch unique identifiers, equipment runtime durations, and cross-process associated node information.
Constraints on Model Access and Configuration from These Data Characteristics
Multi-source heterogeneous data format differences require access configurations to support structured database tables, semi-structured equipment logs, and unstructured wafer map files. Ledger data with different update frequencies requires configured incremental synchronization trigger rules to avoid duplicate synchronization or missed real-time data. Nanoscale high-precision parameters and cross-process hierarchical associated fields require the model to recognize and retain data precision and association relationships to avoid parsing distortion. Long process chains create long context requirements, requiring sufficient context capacity to avoid truncation of key process and exception handling information.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Wafer fabrication ledgers contain multi-process, cross-equipment associated data, and must accommodate complete batch operation records and process parameters to avoid truncation of critical information |
maxTokens | 2000–3000 token | The output length of a single report must cover yield analysis, exception handling summaries, and improvement suggestions to ensure complete output content |
quoteMaxToken | 1500–2000 token | Sufficient process recipe and SPC chart associated data must be recalled as reference basis, while controlling total context length to avoid model overload |
temperature | 0.1–0.3 | Ledger aggregation and report generation must maintain data accuracy, avoiding content generated that deviates from original records |
AIPROXY_API_ENDPOINT | Fill in the dedicated access address provided by the platform | Ensure correct routing of model invocation requests and reduce cross-region access latency |
AIPROXY_API_TOKEN | Fill in the dedicated key assigned by the platform | Verify invocation identity to ensure security of data transmission and model invocation |
Three Common Misconfigurations
- Symptom: Tasks show no progress after switching models and cannot roll back to the original configuration. Cause: Context reset rules after model switching are not properly configured, and residual old model scheduling parameters prevent normal task progression.
- Symptom: Associated analysis content for wafer defect maps is missing from generated reports. Cause: The
visionparameter is not enabled, so the model cannot recognize and parse wafer map data embedded in ledgers. - Symptom: Large model invocation occasionally returns empty responses. Cause:
maxContextis set too high, causing model context overflow, orAIPROXY_API_TOKENconfiguration has expired, causing authentication failure.
How to Confirm Configurations Are Properly Set
- Use the built-in model debugging tool in FastGPT v4.17.0, enter ledger keywords for a single wafer batch, and verify that returned content includes accurate references to corresponding processes and equipment parameters.
- Check model invocation logs to confirm there are no format errors in
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENconfigurations, and no authentication failure or request timeout logs. - Adjust the
temperatureparameter to 0.2, generate two copies of the same ledger report, and verify consistency of core data to confirm no random deviations occur. - Test multi-source data access scenarios to verify that ledger data from MES systems and equipment FDC modules can be correctly recalled and associated for display.
The values provided are common starting points and should be measured against the reader's own samples.